系统工程与电子技术

• 软件、算法与仿真 • 上一篇    下一篇

基于SIFT特征匹配的实时鲁棒视频去抖动系统

於俊1,2, 汪增福1,2,3   

  1. 1. 中国科学技术大学自动化系, 安徽 合肥 230027;
    2. 语音及语言信息处理国家工程实验室, 安徽 合肥 230027;
    3. 中国科学院智能机械研究所, 安徽 合肥 230031
  • 出版日期:2014-02-26 发布日期:2010-01-03

Realtime and robust video stabilization system based on SIFT feature matching

 YU Jun1, 2, WANG Zeng-fu1, 2, 3   

  1. 1. Department of Automation, University of Science & Technology of China, Hefei 230027, China;
    2. National Laboratory of Speech and Language Information Processing, Hefei 230027, China; 
    3. Institute of Intelligent Machine, Chinese Academy of Sciences, Hefei 230031, China
  • Online:2014-02-26 Published:2010-01-03

摘要:

面向视频去抖动领域,提出了一个实时系统。在有效地利用尺度不变特征转换算法的鲁棒特征提取特性和随机采样一致算法的鲁棒拟合特性的基础上,所提系统可以根据运动参数的变化剧烈程度,自动调整低通滤波器的尺寸来确定抖动参数以实现图像补偿,从而有效地避免了过稳和欠稳现象;所提系统将丰富的视频参考信息与图像纹理合成算法结合起来,有效地提高了输出视频的稳定性和完整性。客观实验结果表明,该系统在峰值信噪比和耗时方面具有较好的综合优势。主观实验结果表明,所提系统在消除抖动视频中让人不舒适感方面具有较好的优越性。

Abstract:

In view of video stabilization area, a real time system is proposed. Based on the scale invariant feature transform for robust feature extraction and the random sample consensus algorithm for robust fitting, image compensation is applied according to jittered parameters got by low pass filter with adaptive length according to the extent of video content variation, thus over stabilization and under stabilization are prevented effectively. Stable and complete video is gotten after that each frame is repaired with texture synthesis and abundant reference frames. The object experiment results confirmed the comprehensive advantage in peak signal to noise ratio and time cost of this system. The subjective experiment indicates the suitability of removing non comfort on human visual system after video stabilization with this system.